Update README.md

Author: olegranmo

README.md

@@ -322,9 +322,11 @@ The below figure depicts training and test accuracy for the Convolutional Tsetli
 Test accuracy peaks at 99.50% after 168 epochs and 99.51% after 327 epochs. Further, it climbs quickly in the first epochs, passing 99% already in epoch 2. Training accuracy approaches 100%. 
 </p>
 
+Let us now look at a few example 10x10 bit patterns produced by the Convolutional Tsetlin Machine for MNIST, including valid convolution positions. 
 <p align="center">
   <img width="75%" src="https://github.com/olegranmo/blob/blob/master/MNIST_Clauses_II.png">
 </p>
+Here "0: UL" means "upper left" of the image for digit "0". We can see clearly that "0: UL", "0: UR", "0: LL" and "0: LR" jointly construct the shape "0''. The clauses for the other digits behave similarly, and we thus just illustrate selected patch positions for each digit.
 
 ## Requirements